Reviewed September 24, 2026. This guide is based on public guidance from NIST, the U.S. Federal Trade Commission, and the U.K. Information Commissioner’s Office. It is not a test of every AI product or legal advice.
An AI assistant can be useful without being the right place for every document. The practical privacy question is specific: what information are you about to share, who can access it, what may be retained, and what controls are available? A blanket promise that a tool is “private” is less useful than answers to those four questions.
Follow the data through the workflow
When you paste text, upload a file, connect a mailbox, or grant an agent access to a service, the risk changes. A model might process your prompt, but the product around it may also keep chat history, logs, uploaded files, or connected-account permissions. Settings and retention policies differ by provider and plan. Read the product’s current privacy notice and account controls for the exact service you use; do not assume an enterprise setting applies to a free consumer account.
NIST’s Generative AI Profile gives organizations a framework for mapping and managing risks, including privacy-related risks. It is a risk-management resource, not a certification that any named chatbot is safe. The FTC has also warned AI companies to honor their privacy and confidentiality commitments when handling consumer data. Source: FTC.
A five-minute decision before you paste
- Classify the input. Does it contain passwords, API keys, private messages, customer records, health details, account numbers, or confidential work? If yes, remove or mask them unless the tool is approved for that data and the task truly requires it.
- Check the destination. Confirm the website and account. A browser extension, embedded assistant, and a provider’s own site may have different data practices.
- Read the controls. Look for retention, deletion, model-improvement or training settings, access permissions, and export options. Confirm what each control actually covers.
- Limit the connection. If an agent can access a mailbox, drive, or calendar, grant only the necessary scope and review or revoke it when the task ends. Human review is sensible before it sends, deletes, or publishes anything.
- Check the output. An AI response can repeat information from your prompt. Review any draft before sharing it with someone else or placing it in a public document.
What deletion and rights requests can mean
Deleting a visible conversation and deleting all associated data are not necessarily the same operation. A product may retain information for specified purposes and periods. Consult its policy for what is removed, what is retained, and how to submit a request. Rights vary by location and the circumstances of processing. The U.K. ICO’s guidance on individual rights in AI systems explains how organizations should make requests practical and consider automated decisions; it is not a universal rulebook for every jurisdiction.
What to ask a provider or employer
- Which people or systems can see prompts, files, and logs?
- How long are they kept, and where are deletion and export controls?
- Are inputs used to improve models? Does the answer change by plan or administrator setting?
- What happens to connected accounts if access is revoked?
- Is there a documented process for an accidental disclosure?
Our method: We used primary guidance from NIST, FTC, and ICO. This article does not claim that every AI provider trains on prompts, that all tools collect the same data, or that a particular setting guarantees confidentiality. Before using a service for sensitive work, verify its current terms and your organization’s rules.
